Message ID | 20220420095906.27349-1-mgorman@techsingularity.net (mailing list archive) |
---|---|
Headers | show |
Series | Drain remote per-cpu directly | expand |
On Wed, Apr 20, 2022 at 10:59:00AM +0100, Mel Gorman wrote: > This series has the same intent as Nicolas' series "mm/page_alloc: Remote > per-cpu lists drain support" -- avoid interference of a high priority > task due to a workqueue item draining per-cpu page lists. While many > workloads can tolerate a brief interruption, it may be cause a real-time > task runnning on a NOHZ_FULL CPU to miss a deadline and at minimum, > the draining in non-deterministic. Yeah, the non-deterministic is a problem. I saw the kworker-based draining takes 100+ms(up to 300ms observed) sometimes in alloc_contig_range if CPUs are heavily loaded. I am not sure Nicolas already observed. it's not only problem of per_cpu_pages but it is also lru_pvecs (pagevec) draining. Do we need to introduce similar(allow remote drainning with spin_lock) solution for pagevec? > > Currently an IRQ-safe local_lock protects the page allocator per-cpu lists. > The local_lock on its own prevents migration and the IRQ disabling protects > from corruption due to an interrupt arriving while a page allocation is > in progress. The locking is inherently unsafe for remote access unless > the CPU is hot-removed. > > This series adjusts the locking. A spin-lock is added to struct > per_cpu_pages to protect the list contents while local_lock_irq continues > to prevent migration and IRQ reentry. This allows a remote CPU to safely > drain a remote per-cpu list. > > This series is a partial series. Follow-on work would allow the > local_irq_save to be converted to a local_irq to avoid IRQs being > disabled/enabled in most cases. However, there are enough corner cases > that it deserves a series on its own separated by one kernel release and > the priority right now is to avoid interference of high priority tasks. > > Patch 1 is a cosmetic patch to clarify when page->lru is storing buddy pages > and when it is storing per-cpu pages. > > Patch 2 shrinks per_cpu_pages to make room for a spin lock. Strictly speaking > this is not necessary but it avoids per_cpu_pages consuming another > cache line. > > Patch 3 is a preparation patch to avoid code duplication. > > Patch 4 is a simple micro-optimisation that improves code flow necessary for > a later patch to avoid code duplication. > > Patch 5 uses a spin_lock to protect the per_cpu_pages contents while still > relying on local_lock to prevent migration, stabilise the pcp > lookup and prevent IRQ reentrancy. > > Patch 6 remote drains per-cpu pages directly instead of using a workqueue. > > include/linux/mm_types.h | 5 + > include/linux/mmzone.h | 12 +- > mm/page_alloc.c | 333 ++++++++++++++++++++++++--------------- > 3 files changed, 222 insertions(+), 128 deletions(-) > > -- > 2.34.1 > >
On Wed, Apr 20, 2022 at 2:59 AM Mel Gorman <mgorman@techsingularity.net> wrote: > > This series has the same intent as Nicolas' series "mm/page_alloc: Remote > per-cpu lists drain support" -- avoid interference of a high priority > task due to a workqueue item draining per-cpu page lists. While many > workloads can tolerate a brief interruption, it may be cause a real-time > task runnning on a NOHZ_FULL CPU to miss a deadline and at minimum, > the draining in non-deterministic. > > Currently an IRQ-safe local_lock protects the page allocator per-cpu lists. > The local_lock on its own prevents migration and the IRQ disabling protects > from corruption due to an interrupt arriving while a page allocation is > in progress. The locking is inherently unsafe for remote access unless > the CPU is hot-removed. > > This series adjusts the locking. A spin-lock is added to struct > per_cpu_pages to protect the list contents while local_lock_irq continues > to prevent migration and IRQ reentry. This allows a remote CPU to safely > drain a remote per-cpu list. > > This series is a partial series. Follow-on work would allow the > local_irq_save to be converted to a local_irq to avoid IRQs being > disabled/enabled in most cases. However, there are enough corner cases > that it deserves a series on its own separated by one kernel release and > the priority right now is to avoid interference of high priority tasks. > > Patch 1 is a cosmetic patch to clarify when page->lru is storing buddy pages > and when it is storing per-cpu pages. > > Patch 2 shrinks per_cpu_pages to make room for a spin lock. Strictly speaking > this is not necessary but it avoids per_cpu_pages consuming another > cache line. > > Patch 3 is a preparation patch to avoid code duplication. > > Patch 4 is a simple micro-optimisation that improves code flow necessary for > a later patch to avoid code duplication. > > Patch 5 uses a spin_lock to protect the per_cpu_pages contents while still > relying on local_lock to prevent migration, stabilise the pcp > lookup and prevent IRQ reentrancy. > > Patch 6 remote drains per-cpu pages directly instead of using a workqueue. This quite possibly solves the issue I was trying to fix in https://lore.kernel.org/all/20220225012819.1807147-1-surenb@google.com. I will give it a try and see how it looks. Thanks! > > include/linux/mm_types.h | 5 + > include/linux/mmzone.h | 12 +- > mm/page_alloc.c | 333 ++++++++++++++++++++++++--------------- > 3 files changed, 222 insertions(+), 128 deletions(-) > > -- > 2.34.1 > >
On Mon, Apr 25, 2022 at 7:49 PM Suren Baghdasaryan <surenb@google.com> wrote: > > On Wed, Apr 20, 2022 at 2:59 AM Mel Gorman <mgorman@techsingularity.net> wrote: > > > > This series has the same intent as Nicolas' series "mm/page_alloc: Remote > > per-cpu lists drain support" -- avoid interference of a high priority > > task due to a workqueue item draining per-cpu page lists. While many > > workloads can tolerate a brief interruption, it may be cause a real-time > > task runnning on a NOHZ_FULL CPU to miss a deadline and at minimum, > > the draining in non-deterministic. > > > > Currently an IRQ-safe local_lock protects the page allocator per-cpu lists. > > The local_lock on its own prevents migration and the IRQ disabling protects > > from corruption due to an interrupt arriving while a page allocation is > > in progress. The locking is inherently unsafe for remote access unless > > the CPU is hot-removed. > > > > This series adjusts the locking. A spin-lock is added to struct > > per_cpu_pages to protect the list contents while local_lock_irq continues > > to prevent migration and IRQ reentry. This allows a remote CPU to safely > > drain a remote per-cpu list. > > > > This series is a partial series. Follow-on work would allow the > > local_irq_save to be converted to a local_irq to avoid IRQs being > > disabled/enabled in most cases. However, there are enough corner cases > > that it deserves a series on its own separated by one kernel release and > > the priority right now is to avoid interference of high priority tasks. > > > > Patch 1 is a cosmetic patch to clarify when page->lru is storing buddy pages > > and when it is storing per-cpu pages. > > > > Patch 2 shrinks per_cpu_pages to make room for a spin lock. Strictly speaking > > this is not necessary but it avoids per_cpu_pages consuming another > > cache line. > > > > Patch 3 is a preparation patch to avoid code duplication. > > > > Patch 4 is a simple micro-optimisation that improves code flow necessary for > > a later patch to avoid code duplication. > > > > Patch 5 uses a spin_lock to protect the per_cpu_pages contents while still > > relying on local_lock to prevent migration, stabilise the pcp > > lookup and prevent IRQ reentrancy. > > > > Patch 6 remote drains per-cpu pages directly instead of using a workqueue. > > This quite possibly solves the issue I was trying to fix in > https://lore.kernel.org/all/20220225012819.1807147-1-surenb@google.com. > I will give it a try and see how it looks. My test shows sizable improvement for the worst case drain_all_pages duration. Before the change I caught cases when a drain_local_pages_wq in the workqueue was delayed by 100+ms (not even counting drain_local_pages_wq execution time itself). With this patchset the worst time I was able to record for drain_all_pages duration was 17ms. > Thanks! > > > > > include/linux/mm_types.h | 5 + > > include/linux/mmzone.h | 12 +- > > mm/page_alloc.c | 333 ++++++++++++++++++++++++--------------- > > 3 files changed, 222 insertions(+), 128 deletions(-) > > > > -- > > 2.34.1 > > > >
On Mon, 2022-04-25 at 15:58 -0700, Minchan Kim wrote: > On Wed, Apr 20, 2022 at 10:59:00AM +0100, Mel Gorman wrote: > > This series has the same intent as Nicolas' series "mm/page_alloc: Remote > > per-cpu lists drain support" -- avoid interference of a high priority > > task due to a workqueue item draining per-cpu page lists. While many > > workloads can tolerate a brief interruption, it may be cause a real-time > > task runnning on a NOHZ_FULL CPU to miss a deadline and at minimum, > > the draining in non-deterministic. > > Yeah, the non-deterministic is a problem. I saw the kworker-based draining > takes 100+ms(up to 300ms observed) sometimes in alloc_contig_range if CPUs > are heavily loaded. > > I am not sure Nicolas already observed. it's not only problem of > per_cpu_pages but it is also lru_pvecs (pagevec) draining. > Do we need to introduce similar(allow remote drainning with spin_lock) > solution for pagevec? Yes, I'm aware of the lru problem. I'll start working on it too once we're done with the page allocator (and if no-one beats me to it). That said, I don't know if we can apply the exact same approach, the devil is in the details. :)
On Tue, Apr 26, 2022 at 01:06:12PM +0200, Nicolas Saenz Julienne wrote: > On Mon, 2022-04-25 at 15:58 -0700, Minchan Kim wrote: > > On Wed, Apr 20, 2022 at 10:59:00AM +0100, Mel Gorman wrote: > > > This series has the same intent as Nicolas' series "mm/page_alloc: Remote > > > per-cpu lists drain support" -- avoid interference of a high priority > > > task due to a workqueue item draining per-cpu page lists. While many > > > workloads can tolerate a brief interruption, it may be cause a real-time > > > task runnning on a NOHZ_FULL CPU to miss a deadline and at minimum, > > > the draining in non-deterministic. > > > > Yeah, the non-deterministic is a problem. I saw the kworker-based draining > > takes 100+ms(up to 300ms observed) sometimes in alloc_contig_range if CPUs > > are heavily loaded. > > > > I am not sure Nicolas already observed. it's not only problem of > > per_cpu_pages but it is also lru_pvecs (pagevec) draining. > > Do we need to introduce similar(allow remote drainning with spin_lock) > > solution for pagevec? > > Yes, I'm aware of the lru problem. I'll start working on it too once we're done > with the page allocator (and if no-one beats me to it). That said, I don't know > if we can apply the exact same approach, the devil is in the details. :) I think one necessary step for that (adding spinlock to protect per-CPU lru_pvecs) would be to find a suitable testcase. Mel, do you have anything in mind ? > > -- > Nicolás Sáenz > >